A federal appeals court Tuesday upheld Thomson Reuters' copyright victory over Ross Intelligence, a defunct legal research startup that used Westlaw material to train an AI search tool, Reuters reported. The decision by the Philadelphia-based 3rd U.S. Circuit Court of Appeals is the first by a U.S. appeals court on whether training AI on copyrighted material can qualify as fair use.

What Happened

The 3rd Circuit affirmed a lower-court ruling that rejected Ross's fair-use defense. Its reasoning is under seal. According to the blog Copyright Lately, the public judgment from the panel, led by Judge Tamika Montgomery-Reeves, is one page long and says only that the decision is affirmed. The full opinion is being held back while both sides propose redactions.

Thomson Reuters sued Ross in 2020, accusing it of copying headnotes from Westlaw, the legal research service Thomson Reuters owns. Headnotes are short editorial summaries of points of law in court opinions. Ross used them to train a competing search engine meant to surface relevant case law. Ross shut down its platform in 2021, citing the cost of the litigation.

RelatedNew York City Comptroller Urges Anthropic to Delay IPO Over Safety and Governance Concerns

In February 2025, U.S. Circuit Judge Stephanos Bibas, sitting as a trial judge in Delaware, ruled that Ross had infringed 2,243 headnotes and could not claim fair use. The ruling reversed his own 2023 decision, which had left fair use for a jury to decide. Bibas found that Ross's use was not transformative because it copied the headnotes to build a product that competed with Westlaw. He then allowed Ross to appeal before trial, which is how the case reached the 3rd Circuit.

A Thomson Reuters spokesperson said the company was pleased and that respecting copyright is essential to innovation, according to Reuters. Ross did not respond to requests for comment.

Why It Matters

Dozens of copyright suits brought by authors, news organizations and music labels against AI developers turn on the same defense Ross lost. Until now, those cases have relied on trial-court rulings. In June 2025, two federal judges in San Francisco ruled for Anthropic and Meta on fair use in suits by book authors, though on different reasoning and with caveats about pirated copies and market harm. An appellate decision carries more weight, and the 3rd Circuit's ruling binds lower courts in Pennsylvania, New Jersey and Delaware.

The facts, however, are narrower than in most pending cases. Ross's tool did not generate new text; it retrieved existing court opinions, and it was built to compete head-on with the product whose material it copied. Reuters noted that distinction. AI companies are likely to argue that the ruling says little about large language models trained on broad collections of text, while rights holders will cite it as proof that training use can fail the fair-use test.

What the Sealed Opinion Could Settle

Copyright Lately identified several open questions that the unsealed opinion should answer. One is whether the panel accepted Bibas's reasoning that headnotes are copyrightable because an editor's choices about what to highlight involve creativity, which he likened to a sculptor's work. Another is whether potential harm to a market for AI training data counts under fair use. A third is whether a doctrine that has protected intermediate copying of software code applies to other kinds of material.

What to Watch

The timing of the unsealing is the next milestone, and it depends on how quickly the parties agree on redactions. Lawyers in the generative AI cases will read the opinion for how it treats transformative use and market harm rather than for the outcome. Ross could also ask the full 3rd Circuit or the Supreme Court to review the decision.